Add reliable scoped X public-data source - #8
Conversation
9bc3ac1 to
9df42b0
Compare
|
Repaired and refreshed in signed commit 9df42b0. The Xquik package pin now uses the current Local validation passed for shell syntax, shellcheck, synchronized and drift fixtures, both YAML frontmatter files, Python syntax, the secret scanner smoke test, package and docs reachability, and The repository-wide markdownlint result is unchanged from upstream |
…ASS) Ran the whole suite against fresh sessions (8 blind auto-routing batches). Result: 56 PASS / 0 PARTIAL / 0 FAIL, zero fabricated endpoints/regexes/ sections; all six hard-boundary prompts (B3-B8) refused with real section cites. The run surfaced two doc issues (skills behaved correctly): - 9 expected-behavior cells for prompts 1-33 still cited pre-v2.3 methodology section numbers left dangling by the §7 renumber. Refreshed each to the real anchor, verified against the live SKILL.md headers: #1 §7.5->§7.1, #8 §8.5/§7.5->§8.2/§7.1, #9 §10.1/§11.10/§22->§10/§11/§12, #11 §29->§11(+arsenal §16.14), #12 §31.1/§31.3->§16, #15 arsenal §16.14->email-domain-security §7 (deeper skill now supersedes), #19 methodology §27->§11, #27 §28->§9(+arsenal §29.2), #28 §30.2/§30.3->§15. - Prompt #40 couldn't exercise the $-loss model (it keys on leaked-record counts, not severity counts) — added a 40k-record leaked-credential finding so it hits the $-band, with a note on the $0 tri-state fallback. Self-grade line updated to the 100% run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V1dmSDH4QW2kfveHUfUjd4
|
Thanks for the PR — it actually bundled two unrelated changes, so I've split them. Kept: the Declined: the §46.13 Xquik /
No objection to X/Twitter public-collection methodology if it's vendor-neutral (documenting the official X API + generic public-source techniques, no single paid SDK pinned). A PR in that shape is welcome. Closing this one in favor of #11 for the fix. Appreciate the catch on the exit code. |
) --check printed DRIFT / destination-missing failures but always returned 0, so CI and the smoke driver could not gate on it. Accumulate a CHECK_FAILED flag across all skills and exit 1 when any drifted or was missing; in-sync still exits 0. Cherry-picked from PR #8 (sync fix only; Xquik/x-developer promo dropped). Co-authored-by: Sachin Sharma <elementalsoul@Sachins-MacBook-Pro.local> Co-authored-by: Burak Bayır <8755484+kriptoburak@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
…ASS) Ran the whole suite against fresh sessions (8 blind auto-routing batches). Result: 56 PASS / 0 PARTIAL / 0 FAIL, zero fabricated endpoints/regexes/ sections; all six hard-boundary prompts (B3-B8) refused with real section cites. The run surfaced two doc issues (skills behaved correctly): - 9 expected-behavior cells for prompts 1-33 still cited pre-v2.3 methodology section numbers left dangling by the §7 renumber. Refreshed each to the real anchor, verified against the live SKILL.md headers: #1 §7.5->§7.1, #8 §8.5/§7.5->§8.2/§7.1, #9 §10.1/§11.10/§22->§10/§11/§12, #11 §29->§11(+arsenal §16.14), #12 §31.1/§31.3->§16, #15 arsenal §16.14->email-domain-security §7 (deeper skill now supersedes), #19 methodology §27->§11, #27 §28->§9(+arsenal §29.2), #28 §30.2/§30.3->§15. - Prompt #40 couldn't exercise the $-loss model (it keys on leaked-record counts, not severity counts) — added a 40k-record leaked-credential finding so it hits the $-band, with a note on the $0 tri-state fallback. Self-grade line updated to the 100% run. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01V1dmSDH4QW2kfveHUfUjd4
) --check printed DRIFT / destination-missing failures but always returned 0, so CI and the smoke driver could not gate on it. Accumulate a CHECK_FAILED flag across all skills and exit 1 when any drifted or was missing; in-sync still exits 0. Cherry-picked from PR #8 (sync fix only; Xquik/x-developer promo dropped). Co-authored-by: Sachin Sharma <elementalsoul@Sachins-MacBook-Pro.local> Co-authored-by: Burak Bayır <8755484+kriptoburak@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Description
Adds a scoped X/Twitter public-data source to the offensive OSINT tooling quick-install section. The install command is pinned to the current
x-developer@2.5.3package. The workflow stays limited to authorized public-source collection and does not enable private account access or publishing by default.This PR also fixes
scripts/sync-skill-content.sh --check: detected content drift or a missing destination now returns a failing exit code after checking every configured skill. Missing optional canonical source files retain the script's existing skip behavior.Type of change
Affected skills
osint-methodologyoffensive-osintAffected sections
Checklist
CHANGELOG.mdunder[Unreleased].tests/smoke-test-prompts.md.<type>(<scope>): <subject>.Sample prompt
Expected response
Claude should route to arsenal section 46.13, give the pinned
x-developer@2.5.3install command, link to Xquik docs, and keep the workflow limited to authorized public X/Twitter collection.Independent repository fix
scripts/sync-skill-content.sh --checkpreviously printed drift failures but returned success. The repaired check accumulates failures, still reports every skill, and exits 1 when a canonical source differs from its destination or the destination is missing.Validation
bash -n scripts/sync-skill-content.shshellcheck scripts/sync-skill-content.shSKILL.mdfilessecret_scan.pysmoke testnpm view x-developer@2.5.3 versiongit diff --checkmainThe repository-wide markdownlint command reports the same 686 existing findings on upstream
mainand this branch. GitHub Actions currently requires maintainer approval before its fork workflow can run.